g model article in press - cnrpontrell/pub/mesh_refin2016.pdf · free-fall regime describes the...
TRANSCRIPT
J
De
Ma
b
c
a
ARRAA
KEEDA
1
flamsbtecedflmiisoldis
h1
ARTICLE IN PRESSG ModelOCS-498; No. of Pages 9
Journal of Computational Science xxx (2016) xxx–xxx
Contents lists available at ScienceDirect
Journal of Computational Science
journa l h om epage: www.elsev ier .com/ locate / jocs
ynamic mesh refinement for discrete models of jetlectro-hydrodynamics
arco Lauricellaa, Giuseppe Pontrelli a, Dario Pisignanob,c, Sauro Succia,∗
Istituto per le Applicazioni del Calcolo CNR, Via dei Taurini 19, 00185 Rome, ItalyDipartimento di Matematica e Fisica “Ennio De Giorgi”, University of Salento, via Arnesano, 73100 Lecce, ItalyIstituto Nanoscienze-CNR, Euromediterranean Center for Nanomaterial Modelling and Technology (ECMT), via Arnesano, 73100 Lecce, Italy
r t i c l e i n f o
rticle history:eceived 1 December 2015eceived in revised form 26 April 2016ccepted 4 May 2016
a b s t r a c t
Nowadays, several models of unidimensional fluid jets exploit discrete element methods. In some cases,as for models aiming at describing the electrospinning nanofabrication process of polymer fibers, discreteelement methods suffer a non-constant resolution of the jet representation. We develop a dynamic mesh-refinement method for the numerical study of the electro-hydrodynamic behavior of charged jets using
vailable online xxx
eywords:lectrohydrodynamicslectrospinningiscrete element method
discrete element methods. To this purpose, we import ideas and techniques from the string methodoriginally developed in the framework of free-energy landscape simulations. The mesh-refined discreteelement method is demonstrated for the case of electrospinning applications.
© 2016 Elsevier B.V. All rights reserved.
daptive mesh refinement
. Introduction
Discrete element methods are widely employed to model fluidows in air channels, pipes and several other applications, suchs modeling ink-jet printing processes, electrospinning, spray jets,icro-fluid dynamics in nozzles, etc. [1–5]. In particular, unidimen-
ional jets can be easily modeled as a sequence of discrete elementsy defining a mesh of points, which is used to discretize a con-inuous object (e.g. a liquid body) as a finite sequence of discretelements. The aim of such model is to provide a relatively simpleomputational framework based on particle-like ordinary differ-ntial equations, rather than on the discretization of the partialifferential equations of continuum fluids. Electro-hydrodynamicows, however, are often subject to strong interactions leading toajor deformation of the jet, hence to significant heterogeneities
n the spatial distribution of the discrete particle. The latter, in turn,mply a loss of accuracy of the numerical method, since the mosttretched portions of the jet become highly under-resolved. Onef these cases is the electrospinning process, where a polymericiquid jet is ejected from a nozzle and accelerated toward a con-
Please cite this article in press as: M. Lauricella, et al., Dynamic meshComput. Sci. (2016), http://dx.doi.org/10.1016/j.jocs.2016.05.002
uctive collector by a strong electric field. In this framework, the jets stretched so that its diameter decreases below the micrometer-cale, providing a one-dimensional structure with very high surface
∗ Corresponding author. Tel.: +39 06 4927 0958.E-mail address: [email protected] (S. Succi).
ttp://dx.doi.org/10.1016/j.jocs.2016.05.002877-7503/© 2016 Elsevier B.V. All rights reserved.
area to volume ratio. This intriguing feature of the resulting poly-mer nanofibers spawned several papers [6–11] and books [12–14]focussing on the electrospinning process.
In this framework, pioneering works by the Reneker and Yaringroups were focused on developing ad-hoc discrete element meth-ods for electrospinning, which describe nanofibers as a series ofbeads obeying the equations of Newtonian mechanics [1,15]. Thismodeling approach has gained an important role in predicting theoutcome of electrospinning experiments. In addition, such modelsmight support experimental researchers with a likely starting pointfor calibrating processes in order to save time before subsequentoptimization work.
In these models, the excess charge is distributed to each ele-ment composing the jet representation, and it is static in the frameof reference of the extruded fluid jet. Once the solution surfacetension is overcome by electrical forces at the spinneret, the jetserves as fluid medium to push away the mutually repulsing elec-tric charges from the droplet pending at the nozzle of the apparatus.The main forces affecting the jet dynamics, which are accounted forin such models, include viscoelasticity, surface tension and electro-static interactions with the external field and among excess chargesin the liquid. In particular, the Coulomb repulsion between elec-tric charges triggers bending instabilities in electrically charged
refinement for discrete models of jet electro-hydrodynamics, J.
jets, as demonstrated by Reneker et al. [15]. Despite being oftenneglected, air drag and aerodynamic effects, which may also leadto bending instabilities, have been recently modeled by discreteelement methods [16–18]. Further, several complex viscoelastic
ING ModelJ
2 puta
mssd
aarunpttattafder
astoamlpist
tttraia
rmwtwF
2
mdstmsECtta(t
ARTICLEOCS-498; No. of Pages 9
M. Lauricella et al. / Journal of Com
odels were included in order to simulate viscoelastic Boger fluidolutions [19,20]. Systematic investigations were carried out oneveral simulations parameters: polymer concentration, solutionensity, electric potential, and perturbation frequency [21,22].
Notwithstanding a satisfactory, though generally qualitative,greement with experimental results has been shown [23,24,15],ll these methods suffer the non-constant resolution of the jet rep-esentation, which usually decreases downstream. Indeed, given aniform discretization of the initial polymeric drop pending at theozzle, the initial step of the jet dynamics is characterized by therevalent viscoelastic force, and it leads to a second regime, whenhe longitudinal stress changes under the effect of the applied elec-ric voltage, accelerating the jet toward the collector [15]. Hence,
free-fall regime describes the later jet dynamics. Since the longi-udinal viscoelastic stress along the fiber is usually larger close tohe nozzle, the distance between each discrete element increasess a result of the uniformly accelerated motion, which drives thearthest elements from the spinneret. As a consequence, the jetiscretization close the collector becomes rather coarse to modelfficiently the filament, and the information (position, velocity,adius, stress, etc.) describing the jet is lost downstream.
A refined description of jet is necessary in every instance where high-resolution description of physical quantities is requested,uch as for bending instabilities, varicose instabilities of diame-er, etc. Further, a dense mesh provides a more strict assessmentf the Coulomb repulsion term, which is important to properlyccount the transverse force acting on the jet. Recently, refine-ent procedures were proposed in few works in order to avoid
ow resolution problems downstream [25,26]. Nonetheless, theserocedures exploit only linear interpolations, and are not able to
mpose a uniform mesh of jet representation, which is likely theimplest choice to model properly the entire jet, from the nozzle tohe collector.
Here, we present an algorithm specifically developed to addresshe issue. The aim of the algorithm is to recover a finer jet represen-ation at constant time interval, before the information describinghe jet is scattered downstream, so as to preserve the jet modelingepresentation and make the simulation more realistic. Further, thelgorithm can be used to enforce several types of mesh depend-ng on the physical quantities under investigation, providing andaptive description of the process.
The article is organized as follows. In Section 2, we summa-ize the 3D model for electrospinning, with the set of equations ofotion (EOM) which govern the dynamics of system. In Section 3,e present the algorithm, with the relative step for its implementa-
ion. In Section 4 we report a numerical example of its applicationithin a discrete element method for electrospinning modeling.
inally, conclusions are outlined in Section 5.
. Model
In this work, we apply our algorithm to the electrospinningodel implemented in JETSPIN, an open-source code specifically
eveloped for electrospinning simulation [23], which is brieflyummarized in the following text. The model is an extension ofhe discrete model originally introduced by Reneker et al. [15]. The
ain assumption of such model is that the filament can be rea-onable represented by a series of n discrete elements (jet beads).ach one of this elements is a particle-like labeled by an index ith. Aartesian coordinate system is taken so that the origin is located athe nozzle, and x-axis is pointing perpendicularly toward a collec-
Please cite this article in press as: M. Lauricella, et al., Dynamic meshComput. Sci. (2016), http://dx.doi.org/10.1016/j.jocs.2016.05.002
or where the nanofiber is deposited. The nozzle is also representeds a particle like (nozzle bead) which can moves on the plane y–zfurther details in the following text). Given a fluid jet starting athe nozzle and labeling i = 0 the nozzle bead, we obtain the set of
PRESStional Science xxx (2016) xxx–xxx
beads indexed i = 0, 1, . . ., n, where n denotes the bead of the otherextremity of the filament. Thus, given �ri the position vector of ageneric ith bead, we define the mutual distance between i and i + 1elements
li =∣∣�ri+1 − �ri
∣∣ . (1)
Here, li stands for the length step used for discretizing the fil-ament. Note that the length l is typically larger than the filamentradius, but smaller than the characteristic lengths of other observ-ables of interest (e.g. jet curvature radius). Denoted mi the mass, tthe time, and ��i the velocity vector of ith bead, its acceleration isgiven by Newton’s law
d ��i
dt=
�ftot,i
mi, (2)
where �ftot,i is the total force acting on ith bead, given as a sum ofseveral terms
�ftot,i = �fel,i + �fc,i + �f�e,i + �fst,i. (3)
In last equation, �fel,i stands for the electric force due to the exter-nal electrical potential �0 imposed between the nozzle and thecollector located at distance h along versor �x, and given as
�fel,i = qi�0
h· �x. (4)
Denoted qi the charge of ith bead, the net Coulomb force �fc is
�fc,i =n∑
j = 1
j /= i
qiqj∣∣�rj − �ri
∣∣2· �uij (5)
acting on the ith bead and originated from all the other jth beads.Further, we denote �fst the surface tension force
�fst,i = ki · �(
ai + ai−1
2
)2 ̨ · �ci, (6)
acting on ith bead toward the center of the local curvature (accord-ing to �ci versor) to restore the rectilinear shape of the jet, given ˛the surface tension coefficient, ai the jet radius, and ki the local cur-vature (measured at ith bead). Finally, �f�e stands for the viscoelasticforce
�f�e,i = −�a2i �i · �ti + �a2
i+1�i+1 · �ti+1, (7)
pulling bead i back to i − 1 and toward i + 1, with � the stress and�ti the unit vector pointing bead i from bead i − 1. The polymeric jetis assumed as a viscoelastic Maxwellian liquid so that � evolves intime following the constitutive equation:
d�i
dt= G
li
dlidt
− G
��i, (8)
where G is the elastic modulus, and � is the viscosity of the fluidjet.
Finally, the velocity ��i satisfies the kinematic relation:
d�ri
dt= ��i. (9)
The set of three Eqs. (2), (8) and (9) governs the time evolutionof system.
The nozzle bead with charge q̄0 and x0 = 0 is experiencing uni-form circular motion in order to model fast mechanical oscillationsof the spinneret [21], where we denote A the amplitude of the
refinement for discrete models of jet electro-hydrodynamics, J.
perturbation and ω its frequency.The initial simulation conditions as well as the jet insertion at
the nozzle are briefly described in the following. We assume that allthe electrospinning simulations start with a single jet bead, which
ARTICLE ING ModelJOCS-498; No. of Pages 9
M. Lauricella et al. / Journal of Computat
F(t
igapwE2tjbtfits
ficatseaetnt(
3
(aitplflt(wsfcit(Astb
ig. 1. Sketch of a liquid jet (a) and its modeling representation by discrete elementsb) drawn as circles with connections representing the mutual interactions betweenwo consecutive elements.
s placed at distance lstep from the nozzle along the x axis. The sin-le bead represents a jet with a cross-sectional radius a0, defineds the radius of the filament at the nozzle before the stretchingrocess. The jet bead has an initial velocity �s along the x axishich is impressed by the bulk fluid velocity in the syringe needle.
volving in time the system, this jet bead travels up to a distance·lstep away from the nozzle, and a new jet bead is placed at dis-ance lstep between the two previous bodies (nozzle and previouset bead). The procedure is repeated so that we obtain a series of neads representing the jet. Therefore, the parameter lstep representshe length step which is used to discretize the polymer solutionlament in discrete elements, and, in other words, it is the spa-ial resolution imposed at the nozzle before the stretching processtarts acting.
Each simulation starts with only two bodies: the nozzle bead atxed at x0 = 0 and a single bead modeling an initial jet segment ofross-sectional radius ainit located at distance lstep from the nozzlelong the x axis with an initial velocity �b along the x axis equal tohe bulk fluid velocity in the nozzle. Here, lstep denotes the lengthtep used to discretize the polymer solution filament in discretelements, and, in other words, it is the spatial resolution imposedt the nozzle before the stretching process starts acting. Under theffect of the external electric field the jet bead starts to move towardhe collector, and as soon as it is a distance of 2·lstep away from theozzle, a new jet bead (third body) is placed at distance lstep fromhe nozzle along the straight line joining the two previous bodiesfurther details in Ref. [23]).
. Algorithm
Given a series of discrete elements representing a liquid jetFig. 1), our main objective is to recover a finer representation,s soon as it has become too coarse. As example, if our aim is tonvestigate cross section fluctuations of the nanofiber deposited onhe collector at a length scale of 1 mm, as soon as the stretchingrocess moves two beads of the discretized jet at mutual distance
arger than 1 mm, the representation is too coarse for resolving suchuctuations at the desired length scale. For this purpose, we notehat the jet elements define a natural parameterization of a stringcurve), which starts and ends at two given extreme points. Thus,e borrow and adapt ideas and techniques from elements of the
implified string method originally proposed by Weinan et al. [27]or computing the minimum energy path in an energy landscapeontext. In the original simplified string method, a string is dynam-cally evolved in the free energy landscape in order to identifyhe reactive path connecting two minima through a saddle pointusually the transition state of the process under investigation).
Please cite this article in press as: M. Lauricella, et al., Dynamic meshComput. Sci. (2016), http://dx.doi.org/10.1016/j.jocs.2016.05.002
n initial string is discretized by a uniform mesh (points equallypaced). Then, the string (each point of the mesh) is evolved inime by a steepest descent dynamics and the mesh undergoes thereak of its uniform distribution. At constant time interval, a check
PRESSional Science xxx (2016) xxx–xxx 3
on the mesh uniformity is performed, and, if necessary, the uniformrepresentation of the mesh is reinforced by a cubic spline interpo-lation. Here, our idea is to apply a refinement procedure at constanttime interval during the jet hydrodynamics whenever the mutualdistance between any two consecutive elements representing thejet exceeds a prescribed threshold. Note that the refinement proce-dure should be performed before the jet representation is scatteredby the stretching dynamics. In other words, we wish to preserve themodeling information before it vanishes.
By integrating in time the equations of motion above (see thefield “integrate the system” in Fig. 3), the observable li defined inEq. (1) between two consecutive elements i and i + 1 is found toincrease under the main action of the external electric field. Thealgorithm proceeds in three steps as follows.
3.1. First step
At regular interval time tref, we compute the mutual distanceli between all the elements of the actual jet representation. Notethat the parameter tref should be set in order to perform the checkbefore the information of the jet representation is lost due to thestretching. Starting from the nozzle (i = 0), all the mutual distanceli are tested for checking whether the length li is larger than a pre-scribed threshold length l̃. In particular, we mark by s ∈ [0, . . ., n] thebead closest to the nozzle, whose distance ls to the next bead s + 1 islarger than threshold l̃ (see Fig. 2). If li < l̃ for any i, the simulationproceeds normally, otherwise we move to the second step.
3.2. Second step
We introduce the curvilinear coordinate � ∈ [0, 1] to parame-terize the jet path, where � = 0 identifies the nozzle, and � = 1 the jetat the collector. For each bead we compute its respective �i valueby using the formula:
�i = 1lpath
i∑k=1
lk (10)
where lpath =∑n
k=1lk is the jet path length from the nozzle to thecollector. In other words, we calculate the arc-lengths correspond-ing to the current parameterization of the string (polymer solutionjet). The set of �i values represent the mesh used to build a cubicspline. Denoted generically by y one of the main quantities describ-ing the jet, the data yi = y (�i) are tabulated and used in order tocompute the coefficients of the spline. Note that several piecewisecubic functions are available in the literature [28]. Here, we usethe Akima’s algorithm [29] in order to compute spline coefficients,since it was demonstrated that other natural cubic splines couldoscillate, whenever the tabulated values change quickly [28]. Thecubic coefficients are computed for all the main quantities describ-ing the jet beads (positions, fiber radius, stress, velocities).
3.3. Third step
The filament is refined starting from the bead with index s up tothe last bead (the farthest bead from the nozzle i = n), and only inthis part of the string a uniform discretization of arc-length equalto l̃ is imposed to the representation. In fact, we use a uniformdiscretization only for the sake of simplicity, and this choice is
refinement for discrete models of jet electro-hydrodynamics, J.
not mandatory. The interpolation is performed by using the splinecoefficients previously computed in the second step. Starting fromthe bead with index s, we enforce the equal arc-length parameter-ization (uniform parameterization) of the filament by imposing all
Please cite this article in press as: M. Lauricella, et al., Dynamic mesh refinement for discrete models of jet electro-hydrodynamics, J.Comput. Sci. (2016), http://dx.doi.org/10.1016/j.jocs.2016.05.002
ARTICLE IN PRESSG ModelJOCS-498; No. of Pages 9
4 M. Lauricella et al. / Journal of Computational Science xxx (2016) xxx–xxx
Fig. 2. Diagram (not in scale) showing the actual jet representation (in blue), and therefined representation with uniform discretization of arc-length equal to l̃ (in red)starting from the bead with index s. We report the length step lstep used to discretizethe jet at the nozzle before the stretching. We also highlight the threshold distancel̃ used to discern the bead where the spline interpolation starts (bead with indexs). Note the string is interpolated by using a constant arc-length equal to l̃. (Forinterpretation of the references to color in this figure legend, the reader is referredto the web version of this article.)
Fig. 3. Flow chart of the algorithm. The three steps of the procedure are highlightedin different colors. (For interpretation of the references to color in this figure legend,the reader is referred to the web version of this article.)
IN PRESSG ModelJ
putational Science xxx (2016) xxx–xxx 5
ta
n
w⌈
�
wiztainTaqtasj
itppos
4
fifaiaclins
crfanetI
TSip
Fig. 4. Mean value of the resolving power 〈R (�)〉 computed as function of the curvi-linear coordinate � over the simulation time in stationary regime for the case I (blue
ARTICLEOCS-498; No. of Pages 9
M. Lauricella et al. / Journal of Com
he arc-lengths of the elements to be equal to l̃ (see Fig. 2). First ofll, we define the number of beads in the new representation as:
∗ = s +⌈∑n
k=s+1lk
l̃
⌉, (11)
here the symbol � · � denotes the ceiling function. Denoted nref =∑n
k=s+1lk
l̃
⌉, the new mesh �∗
iis obtained as:
∗i =
⎧⎨⎩
�i if i ≤ s
�∗i−1 + 1 − �s
nrefif i > s
(12)
here �s is the value of the parameter � at the bead s. Note that its still �* ∈ [0, 1], and the parameter �* has the same values fromero (the nozzle) up to the bead s. In this way, we do not applyhe refinement procedure close to the nozzle, where new beads aredded to the system, in order to not perturb the mechanism of beadnjection (for the injection mechanism see Ref [23]). Finally, theew values y
(�∗
i
)are computed for all the �∗
iby spline interpolation.
he procedure is performed for y to be positions, fiber radius, stressnd velocities of the jet beads in order to provide the correspondinguantities in the new mesh �∗
i. Then, the new system is evolved in
ime following Eq. (1). It is worth stressing that, in principle, it islso possible to define an adaptive mesh of the string by using auitable weight function of a generic observable O of the string (e.g.et radius).
A summarizing flow chart of the algorithm has been sketchedn Fig. 3. It is worth stressing that, notwithstanding the length ofhe curve in Eq. (10) is computed by linear interpolation, the aboverocedure of reparameterization has the accuracy of a third-degreeolynomial, since we use Akima’s cubic spline for the interpolationf the curve. This is fine, since we care mostly about preserving thetring accuracy than computing carefully its arc-length.
. Applications on electrospinning
The dynamic refinement method was implemented in a modi-ed version of JETSPIN, an open-source code specifically developed
or electrospinning simulations [23]. The JETSPIN code delivers discrete element model, which bases on the model originallyntroduced by Reneker et al. [15]. As a test case, we perform
simulation for a polyvinylpyrrolidone (PVP) solution, which isommonly used in electrospinning experiments [10,30]. The simu-ations are carried out by using the simulation parameters reportedn Ref. [23], modeled on the experimental data reported by Monti-aro et al. [11]. For convenience, all the simulation parameters areummarized in Table 1.
We now consider two different simulations (in the followingase I and case II), the second incorporating the mesh-refined algo-ithm. The simulations were integrated in time for 0.2 s by theourth order Runge–Kutta method with a time step equal to 10−8 s,nd the jet beads were inserted at a distance lstep = 0.02 cm from the
Please cite this article in press as: M. Lauricella, et al., Dynamic meshComput. Sci. (2016), http://dx.doi.org/10.1016/j.jocs.2016.05.002
ozzle. In the second simulation the dynamic refinement is appliedvery tref = 0.001 s. During the refinement procedure we impose allhe arc-lengths l̃ equal to 0.2 cm for the new parameterization.n both simulations, we distinguish two different regimes of the
able 1imulation parameters for the simulations of PVP solution electrified jets. The headings unitial fluid velocity at the nozzle, ˛: surface tension, �: viscosity, G: elastic modulus, V0
erturbation at the nozzle.
(kg/m3) q (C/L) a0 (cm) �s (cm/s) ̨ (mN/m)
840 2.8 · 10−7 5 · 10−3 0.28 21.1
line) and the case II (red dashed line) introduced in the text, respectively. (For inter-pretation of the references to color in this figure legend, the reader is referred to theweb version of this article.)
observables describing the electrospinning process. In the first, theobservables follow an initial drift when the jet has not reached thecollector yet. Then, all the observables fluctuate around a constantmean value (stationary regime) after the jet touches the collector.Here, we focus our attention on the latter regime. Given the seriesof n beads representing the jet, we define the observable:
Ri = lstep
li(13)
where li is the mutual distance between i and i + 1 elements, andlstep the length step. We stress that R ∈ (0, 1], and, in particular, R0is equal to 1 at the nozzle, since l0 is imposed equal to the lengthstep lstep, which is used to discretize the jet before stretching, andit works as a resolving power indicator of the actual jet represen-tation at a specific point of the filament. Note that in the following,we use the notation R(�) instead of Ri just for the sake of simplicity,exploiting that, given a configuration, a unique value of � is asso-ciated to each ith bead, where the observable Ri is computed. InFig. 4 we show the mean value 〈R (�)〉 computed as function of thecurvilinear coordinate � over the entire stationary regime. Here, weobserve a quick decrease of 〈R (�)〉 for the case I. For instance, theresolving power R is already 1/40 of its initial value at � equal to 0.1,and it continues to decrease up to 1/200 of the initial R value closethe collector at curvilinear coordinate � = 1. On the other hand, theresolving power R for the caseII is larger than 0.075 for any value of�, showing that the algorithm is efficient in preventing too coarsediscretization of the jet. Thus, we obtain the finer representation ofthe polymeric filament displayed in Fig. 5, where the beads belong-ing to the new discretization are colored in red, and the grains ofthe old one are in blue.
The higher resolving power R of case II safeguards a finer dis-crete representation of the jet which is a continuous object. This is
refinement for discrete models of jet electro-hydrodynamics, J.
sed are as follows: : density, q: charge density, a0: fiber radius at the nozzle, �s:: applied voltage bias, ω: frequency of perturbation at the nozzle, A: amplitude of
� (Pa s) G (Pa) V0 (kV) ω (s−1) A (cm)
2.0 5 · 104 9.0 104 10−3
particularly useful in modeling varicosity [31]. As an example, wereport an electrospinning simulation of a varicose jet in order totest our refinement method in this special framework. Here, vari-cose jet stands for a filament with rapid fluctuations in its cross
ARTICLE IN PRESSG ModelJOCS-498; No. of Pages 9
6 M. Lauricella et al. / Journal of Computational Science xxx (2016) xxx–xxx
Fig. 5. Two snapshots showing the jet representation in the case I (in blue, left panel), and in the case II (in red, right panel). The refined procedure was carried out in thec pretav
sawTcFcrsilmo
tsmiidmimc
�f
wlsIibtscwa
ase II imposing a uniform discretization with arc-length l̃ equal to 0.2 cm. (For interersion of this article.)
ection radius. The varicosity was artificially induced by insertingt the nozzle a bead every ten beads with an extra 10% of mass,hile all the other parameters are kept the same as in Table 1.
his extra mass models a solid impurity, which triggers the vari-osity along its jet path from the nozzle toward the collector. Inig. 6 we report a short segment of the filament deposited at theollector for the two cases. Here, we note that in case II the meshefinement is capable of representing the fluctuation of the crossection along the nanofiber. On the other hand, such informations completely missed in the case I, where the filament results atower resolution. This highlights the utility of the dynamic refine-
ent procedure, whenever our aim is to model nanofiber featuresf small length-scale.
Next, we offer few comments on the parameter l̃ which defineshe resolution parameter used in the spline interpolation of thetring. It was observed by Yarin et al. [1] that the discrete ele-ent modeling of a fiber by 0-dimensional point particles (beads)
mplies mathematical inconsistencies, as the discretization densityncreases. Indeed, the modeled fiber tends to an infinitely thin, one-imensional continuous object, invalidating thus the discrete beadodel. The repulsive Coulomb force is the term incurring in such
ssue, since the electrostatic force between two consecutive jet seg-ents of length lstep and with their junction point at �p value in
urvilinear coordinate is computed by the double integral
1→2 = 14�ε0
∫ �p
�p−lt
d�1
∫ �p+lt
�p
d�2q2
l∣∣�r (�2) − �r (�1)∣∣3
(�r(�2) − �r(�1)) (14)
here �r (�) denotes the position vector along the string at curvi-inear coordinate �, and ql is the linear charge density. It is easilyeen that the integral diverges as the parameter l̃ tends to zero.n an analytical study Yarin et al. [1] addressed this issue by argu-ng that charges are not distributed on the centerline of the fiber,ut they lies rather on its outer shell of circular shape. Thus,he problem is recovered by accounting for the actual electro-
Please cite this article in press as: M. Lauricella, et al., Dynamic meshComput. Sci. (2016), http://dx.doi.org/10.1016/j.jocs.2016.05.002
tatic form factors between two interacting circular sections of aharged filament. In this context, several analytical approximationsere developed [1,26,25] for computing the local electric force
cting between two contiguous rings representing jet elements.
tion of the references to color in this figure legend, the reader is referred to the web
We recommend to use one of the mentioned analytical approxi-mations, if really low values of l̃ are employed during the dynamicsrefinement procedure. However, a reasonable value of l̃ can beassessed in order to maintain a fine jet representation and avoid-ing, at the same time, mathematical inconsistencies. In particular,the Coulomb integral for the electrostatic repulsion between twointeracting rings can be numerically computed and compared withEq. (14) in order to estimate the error introduced by the center-line approximation at the given value of l̃. A conservative valueof l̃ larger than the length step lstep introduces a negligible errordue to the centerline approximation. In this work, we impose avalue of l̃ which is an order of magnitude larger than length steplstep. In order to justify the last statement, we report in Fig. 7 snap-shots of several simulations performed with three distinct valuesof l̃ equal to 0.2 cm, 0.4 cm, 0.8 cm, and a fourth simulation withoutthe dynamic refinement activated at time t equal to 0.009 s, just0.001 s before the jet touches the ground. Here, it is evident as thejet topology is substantially equivalent, confirming that possibleerrors introduced by our approximation for the Coulomb force ofEq. (5) at our operative values of l̃ are negligible.
Furthermore, it is worth to stress that a small value of l̃ alsoimplies a larger computational cost of the simulation. Indeed, thecomputation of the repulsive Coulomb force is a n-body problem,which scales as O
(n2
)with n denoting the number of discrete ele-
ments. As example, we report in Table 2 the CPU wall-clock timeof the three electrospinning simulations, with the dynamic refine-ment procedure and l̃ equal to 0.2 cm, 0.4 cm, and 0.8 cm, previouslymentioned. Along with these data, the computational cost of a sim-ulation without dynamic refinement is also shown. For the sake ofcompleteness, we also report in Table 2 the parallel efficiency �,which is defined as � = Ts/(Tp * nproc) with Ts and Tp the CPU wall-clock time for the same job executed in serial and parallel mode,and nproc the number of CPUs involved in the parallel run. Here,we note that the increase of effective CPU time is dependent on
refinement for discrete models of jet electro-hydrodynamics, J.
the average number of discrete elements n, as expected. In partic-ular, we note that the simulation with dynamic refinement impliesa higher CPU cost than a classical simulation (case I). In Ref. [25],Kowalewski et al. addressed the problem by using the hierarchical
Please cite this article in press as: M. Lauricella, et al., Dynamic mesh refinement for discrete models of jet electro-hydrodynamics, J.Comput. Sci. (2016), http://dx.doi.org/10.1016/j.jocs.2016.05.002
ARTICLE IN PRESSG ModelJOCS-498; No. of Pages 9
M. Lauricella et al. / Journal of Computational Science xxx (2016) xxx–xxx 7
Fig. 6. Snapshots of the nanofiber segments deposited on the collector for the case I on top left (a) and for the case II in red dashed line on top right (b). In black arrows wehighlight varicose defects along the filaments. We also report the fiber radius profile for the case I on bottom left (c), and for the case II on bottom right (d). The cross sectionradius was multiplied by three orders of magnitude in order to highlight varicosity. (For interpretation of the references to color in this figure legend, the reader is referredto the web version of this article.)
Fig. 7. Snapshots of the liquid jet taken at time t equal to 0.009 s for four different simulations with identical simulation parameters apart the value of l̃. Starting from left,denoted (a) simulation without dynamic refinement activated, (b)–(d) three different simulations with dynamic refinement activated and l̃ equal to 0.2 cm, 0.4 cm and 0.8 cm,respectively. Note in figure we have added two lights along the bisector of the plane y–z in order to emphasize the depth of field.
Table 2We report the CPU wall-clock time in seconds which is needed to run several simulations with and without dynamic refinement at different values of l̃. We report also theaverage number of beads used to discretize the jet. The benchmark was executed on a node of 2 × 12 core processors made of 2.4 GHz Intel Ivy Bridge cores.
Dynamic refinement l̃ (cm) # of beads # of CPUs CPU wall clock time (s) �*
No 150 24 8787 0.26Yes 0.8 267 24 12,718 0.44Yes 0.4 509 24 24,726 0.62Yes 0.2 1123 24 75,793 0.80
* It is worth to stress that the parallel efficiency � increases with the number of beads, since the communication latency cost plays a larger role as the simulation involvesa small number of beads, and the computational work is consequently not well distributed over all the CPU cores.
ING ModelJ
8 puta
ffmaacd
5
aoerfltsswtetepc
gjimpe
A
fSm
R
[
[
[
[
[
[
[
[
[
[
[
[
[
[
[
[
[
[
[
[
[
[
[
[
analysis with application in biofluid dynamics, multiscalearterial flows, biological processes modeling, drug deliv-ery. Over the years, he has worked on many researchprojects and developed mathematical models and algo-
ARTICLEOCS-498; No. of Pages 9
M. Lauricella et al. / Journal of Com
orce calculation algorithm (treecode) for computing the Coulomborce, which complexity scales as O (n log n) [32]. The fast multiple
ethod (FMM) [33] can be also used to handle long-range inter-ctions with a high computational efficiency, which theoreticallychieves O (n) operation count. The treecode and FMM are the bestandidate to overcome this issue, and its implementation in ourynamic refinement code will be considered in future works.
. Conclusions
We have presented a dynamic mesh refinement method forddressing low resolution problems in discrete element modelsf jet hydrodynamics. In particular, we have shown by practicalxamples that the proposed algorithm is able to recover a finerepresentation by enforcing a uniform arc-length discretization of auid jet at constant time interval, before the information describinghe modeled object is scattered downstream. Furthermore, the pre-erved jet representation is able to model short-range phenomena,uch as varicosity, providing more realistic simulations. In thisork, we have described the basic structure of the algorithm and
he main steps for its implementation in discrete element mod-ls of the electrospinning process. In addition, we have discussedhe effect of low values of the resolution parameter l̃ on discretelement models and on the relative computational costs, so as torovide general guidelines for using the dynamics refinement pro-edure in electrospinning models.
In summary, the presented algorithm may be used for investi-ating complex short-range phenomena of unidimensional liquidets as well as to improve the accuracy of discrete element modelsn representing electro-hydrodynamic processes. Several existing
odels might benefit of dynamic refinement in order to better sup-ort experiments, and provide useful insights for enhancing thefficiency of several processes involving jet hydrodynamics.
cknowledgements
The research leading to these results has received fundingrom the European Research Council under the European Union’seventh Framework Programme (FP/2007–2013)/ERC Grant Agree-ent n. 306357 (“NANO-JETS”).
eferences
[1] A.L. Yarin, S. Koombhongse, D.H. Reneker, Taylor cone and jetting from liquiddroplets in electrospinning of nanofibers, J. Appl. Phys. 90 (9) (2001)4836–4846.
[2] J.Q. Feng, A general fluid dynamic analysis of drop ejection indrop-on-demand ink jet devices, J. Imaging Sci. Technol. 46 (5) (2002)398–408.
[3] K. Apostolou, A. Hrymak, Discrete element simulation of liquid-particle flows,Comput. Chem. Eng. 32 (4) (2008) 841–856.
[4] H. Wijshoff, Structure-and fluid-dynamics in Piezo Inkjet Printheads,University of Twente, 2008.
[5] G. Dorr, J. Hanan, S. Adkins, A. Hewitt, C. O’Donnell, B. Noller, Spray depositionon plant surfaces: a modelling approach, Funct. Plant Biol. 35 (10) (2008)988–996.
[6] D.H. Reneker, I. Chun, Nanometre diameter fibres of polymer, produced byelectrospinning, Nanotechnology 7 (3) (1996) 216–223.
[7] D. Li, Y. Wang, Y. Xia, Electrospinning nanofibers as uniaxially aligned arraysand layer-by-layer stacked films, Adv. Mater. 16 (4) (2004) 361–366.
[8] C.P. Carroll, E. Zhmayev, V. Kalra, Y.L. Joo, Nanofibers from electrically drivenviscoelastic jets: modeling and experiments, Korea-Aust. Rheol. J. 20 (2008)153–164.
[9] C. Luo, S.D. Stoyanov, E. Stride, E. Pelan, M. Edirisinghe, Electrospinning versusfibre production methods: from specifics to technological convergence, Chem.Soc. Rev. 41 (13) (2012) 4708–4735.
10] L. Persano, A. Camposeo, C. Tekmen, D. Pisignano, Industrial upscaling of
Please cite this article in press as: M. Lauricella, et al., Dynamic meshComput. Sci. (2016), http://dx.doi.org/10.1016/j.jocs.2016.05.002
electrospinning and applications of polymer nanofibers: a review, Macromol.Mater. Eng. 298 (5) (2013) 504–520.
11] M. Montinaro, V. Fasano, M. Moffa, A. Camposeo, L. Persano, M. Lauricella, S.Succi, D. Pisignano, Sub-ms dynamics of the instability onset ofelectrospinning, Soft Matter 11 (2015) 3424–3431.
PRESStional Science xxx (2016) xxx–xxx
12] S. Ramakrishna, K. Fujihara, W.-E. Teo, T.-C. Lim, Z. Ma, An Introduction toElectrospinning and Nanofibers, vol. 90, World Scientific, 2005.
13] D. Pisignano, Polymer Nanofibers: Building Blocks for Nanotechnology, RoyalSociety of Chemistry, 2013.
14] J.H. Wendorff, S. Agarwal, A. Greiner, Electrospinning: Materials, Processing,and Applications, John Wiley & Sons, 2012.
15] D.H. Reneker, A.L. Yarin, H. Fong, S. Koombhongse, Bending instability ofelectrically charged liquid jets of polymer solutions in electrospinning, J. Appl.Phys. 87 (9) (2000) 4531–4547.
16] M. Lauricella, D. Pisignano, S. Succi, Three-dimensional model forelectrospinning processes in controlled gas counterflow, J. Phys. Chem. A(2016), http://dx.doi.org/10.1021/acs.jpca.5b12450 (article ASAP).
17] M. Lauricella, G. Pontrelli, I. Coluzza, D. Pisignano, S. Succi, Different regimesof the uniaxial elongation of electrically charged viscoelastic jets due todissipative air drag, Mech. Res. Commun. 69 (2015) 97–102.
18] M. Lauricella, G. Pontrelli, D. Pisignano, S. Succi, Nonlinear Langevin model forthe early-stage dynamics of electrospinning jets, Mol. Phys. 113 (17–18)(2015) 2435–2441.
19] G. Pontrelli, D. Gentili, I. Coluzza, D. Pisignano, S. Succi, Effects of non-linearrheology on the electrospinning process: a model study, Mech. Res. Commun.61 (2014) 41–46.
20] C.P. Carroll, Y.L. Joo, Electrospinning of viscoelastic Boger fluids: modeling andexperiments, Phys. Fluids 18 (5) (2006) 053102.
21] I. Coluzza, D. Pisignano, D. Gentili, G. Pontrelli, S. Succi, Ultrathin fibers fromelectrospinning experiments under driven fast-oscillating perturbations,Phys. Rev. Appl. 2 (5) (2014) 054011.
22] C. Thompson, G. Chase, A. Yarin, D. Reneker, Effects of parameters onnanofiber diameter determined from electrospinning model, Polymer 48 (23)(2007) 6913–6922.
23] M. Lauricella, G. Pontrelli, I. Coluzza, D. Pisignano, S. Succi, Jetspin: aspecific-purpose open-source software for simulations of nanofiberelectrospinning, Comput. Phys. Commun. 197 (2015) 227–238.
24] C.P. Carroll, Y.L. Joo, Discretized modeling of electrically driven viscoelasticjets in the initial stage of electrospinning, J. Appl. Phys. 109 (9) (2011) 094315.
25] T.A. Kowalewski, S. Barral, T. Kowalczyk, Modeling Electrospinning ofNanofibers, Springer, 2009.
26] T. Kowalewski, S. Nski, S. Barral, Experiments and modelling ofelectrospinning process, Tech. Sci. 53 (4) (2005) 385–394.
27] E. Weinan, W. Ren, E. Vanden-Eijnden, Simplified and improved stringmethod for computing the minimum energy paths in barrier-crossing events,J. Chem. Phys. 126 (16) (2007) 164103.
28] C. de Boor, A Practical Guide to Splines, Applied Mathematical Sciences,Springer, New York, 2001.
29] H. Akima, A new method of interpolation and smooth curve fitting based onlocal procedures, J. ACM 17 (4) (1970) 589–602.
30] D.H. Reneker, A.L. Yarin, Electrospinning jets and polymer nanofibers,Polymer 49 (10) (2008) 2387–2425.
31] W. Yang, H. Duan, C. Li, W. Deng, Crossover of varicose and whippinginstabilities in electrified microjets, Phys. Rev. Lett. 112 (5) (2014)054501.
32] J. Barnes, P. Hut, A hierarchical o (n log n) force-calculation algorithm, Nature324 (1986) 446–449.
33] L. Greengard, V. Rokhlin, A fast algorithm for particle simulations, J. Comput.Phys. 73 (2) (1987) 325–348.
Marco Lauricella, PhD in physics (University College ofDublin), is a post doc fellow at CNR-IAC, and memberof the Complex Systems in Fluid Dynamics and Biologygroup working in the field of electrospinning of polymernanofibers. His expertise is in atomistic computationalmethods, both quantum and classical. He investigatedrelativistic effects on molecular dissociation energies, elu-cidating the chemical stability of many compounds ofindustrial interest. In the framework of statistical physics,he studied rare events via advanced sampling methods.His research interests include transition phase processes,and particularly nucleation phenomena of clathrate com-pounds. In this framework he also developed a new cluster
analysis algorithm for the automatic recognition of clathrate structures. In 2014 hestarted to work on models of polymer nanofibers and electrospinning processes, inthe framework of the ERC NANO-JETS project.
Giuseppe Pontrelli is senior researcher at Istituto per leApplicazioni del Calcolo of CNR in Rome. His researchinterests include applied mathematics and numerical
refinement for discrete models of jet electro-hydrodynamics, J.
rithms for biology and biomedicine.
ING ModelJ
putat
FHCaHPAitc
ARTICLEOCS-498; No. of Pages 9
M. Lauricella et al. / Journal of Com
Dario Pisignano, PhD in Physics, is Associate Profes-sor at the University of Salento. He coordinates aninterdisciplinary group working on Nanotechnologieson Soft Matter at the Department of Mathematicsand Physics “Ennio De Giorgi” of the University ofSalento and at the National Nanotechnology Laboratory(CNR-Nano). His research activity is focused on the devel-opment of advanced polymer processing and lithographicapproaches, electrospinning, and soft lithographies, forrealizing micro-nanostructured devices, organic lasersand functional systems. He has been International Evalua-tor for projects for the National Science Foundation (NSF)U.S.A., the Austrian Science Fund (FWF), the Czech Science
oundation (GACR) and the Italian Ministry of Education, University and Research.e has been member of the International Advisory Committee of the “Internationalonference on Electrospinning” and of the Optical Materials, Fabrication and Char-cterization Committee of the “Conference on Lasers and Electro-Optics” (CLEO).e serves as Referee of the main international journals in the fields of Solid-State
Please cite this article in press as: M. Lauricella, et al., Dynamic meshComput. Sci. (2016), http://dx.doi.org/10.1016/j.jocs.2016.05.002
hysics, Nanotechnology and Materials Science, including Physical Review Letters,ngewandte Chemie International Edition, Advanced Materials, and Laser & Photon-
cs Reviews. He authored about 170 papers and five patents, and has been awardedhe “Young researcher” Prize at the National INFM-Meeting, the International Obdu-at Prize for Nanoimprinting Lithography, the National Award “Future in Research”
PRESSional Science xxx (2016) xxx–xxx 9
of the Ministry of Education, University and Research, the “Sergio Panizza” Awardfor Optoelectronics and Photonics (Italian Physics Society, SIF), and a Prize by theErwin Schrödinger Society for Nanosciences. He is the recipient of the NANO-JETSERC Ideas Starting Grant.
Sauro Succi, PhD in physics (Ecole Polytechnique Federalede Lausanne) is Research Director at CNR-IAC where heleads the Complex Systems in Fluid Dynamics and Biologygroup working on NANO-JETS modeling activities, GuestProfessor at ETH, Zurich, and affiliated with the PhysicsDepartment of Harvard University and External SeniorFellow of the Freiburg Institute of Advanced Studies. Hehas been visiting and guest scientist at ETH Zurich, Yale,and Harvard Universities. He is author of nearly 300 papersin international scientific journals. His research activityis focused on computational physics and mathematicalmodeling of complex systems and process, which willprovide NANO-JETS with an extra added-value in terms of
refinement for discrete models of jet electro-hydrodynamics, J.
reducing the risks connected to highly innovative simulations of electrified jets. Heis the recipient of numerous international awards, such the Alexander von Hum-boldt Award in Physics, the Killam Award of the University of Calgary, and theRahman Chair of the Indian Academy of Sciences. Dr. Succi is an elected Fellowof the American Physical Society and features in the list of the Top Italian Scientists.